Reviews: Revisiting (\epsilon, \gamma, \tau) -similarity learning for domain adaptation

This paper is a theoretical look at domain adaptation / transfer learning problems through the lenses of similarity learning. The authors have extended an already established similarity learning theoretical framework to cases where the training and testing distributions differ. The authors rigorously prove the following in this paper: - A (\epsilon,\gamma)-good similarity for a problem in a source domain is also is an (\epsilon \epsilon', \gamma)-good similarity in a target domain, assuming the same landmark distribution on both the source and the target. In this case, the problem in the target domain becomes (\epsilon \epsilon' \epsilon'', \gamma) good. Both \epsilon' and \epsilon'' are formally derived in the paper.